|
It seems (to me) like AI is not really in the ballpark of your typical CS subjects such as Compilers, Databases, Computer Architecture, OS, etc. These often require you to know the computer itself on a very low-level, while AI may (or in many cases), may not. I can see that AI is still a study of computation/algorithms, but if you generalize the definition of Computer Science to be like that, then isn't everything a study of a computation or an algorithm in some way or another? (physics, biology, etc). AI also seems like it is both different enough (from the low-level Systems topics), and diverse enough as a field to have its own branch. Then why is it always jumbled together with Computer Science at many universities? |
|
Unfortunately, its an accident of history - namely how universities/departments are setup. I suspect it also has to do with the fact that CS depts have deep pockets and can perhaps more easily get grants than say a psychology dept. Personally I feel it (and machine learning) should actually be a branch of engineering ( cf signal processing, image processing, robotics). The problem is that most CS schools do not teach the "engineering" maths required to handle many areas of AI. I have the same problem with psychology departments - which you are supposed to study [reverse-engineer] perception, cognition etc, without the engineering math/computing skills to actually develop a working model of any of these processes. |
|
The field of AI is considered to be a multi- or interdisciplinary field by most researchers in the field. As you and others have rightly pointed out it touches on Psychology, Philosophy, Linguistics (sometimes), Biology (sometimes), Mathematics (and through that sometimes Physics) and Computer Science. The Computer Science perspective of Artificial Intelligence focuses on the development of algorithms and systems that can accomplish tasks that "require intelligence". Other fields are (at times) more interested in the cognitive process rather than the outcome. While they are less likely to call their courses "Artificial Intelligence" there remains significant interest and research in modeling human perception/cognition/action behavior using computational models. Early proponents did define computer science as the study of computation (I'm blanking on the quote -- if someone could fill it in, i'd definitely appreciate it). This makes computational biology, physics, chemistry, etc. all related partially overlapping with computer science. And I don't think that's wrong. I'd rather question the OP definition of "core" computer science as Compilers, Databases, Computer Architecture, OS. These are necessary concepts, but overly limiting; why not include Algorithms and Data Structures? Once you acknowledge these two as core CS components (which most do) you open up the application of these to and development of these for a variety of applications including those tasks that "require intelligence" to accomplish. |
|
I'd add that there are a number of people in other disciplines that are interested in some of the phenomenological/philosophical concepts that those in Artificial Intelligence study. Here are some examples: Cognitive Psychology (from Psychology), Phenomenology (from Philosophy), Developmental Robotics (from Mechanical Engineering). In my opinion, Artificial Intelligence (as a collection of fields) will have to create a paradigm changing discovery in order to 'become its own field' (to adopt the Popper and Kuhn approach to the history of science). Cheers |
|
AI requires a strong background in either electronics or computer programming to do well. While there's a lot of overlap between other fields (mathematics and linguistics are probably the two largest), it isn't imperative that you be particularly skilled in either linguistics or mathematics to be skilled with machine learning algorithms. On the other hand, I would argue that you cannot be very successful with AI unless you're a reasonably decent coder; or if you're into AI from the robotics end of it, electronics. There's plenty of theory angles on AI: computational linguistics, computational ethics, statistical learning theory, etc. While I agree computers are really the only practical way of implementing an AI, I wouldn't say you have to actually implement something to make a contribution to the field. Though if you've found your way to this forum, you're likely a coder.
(Feb 08 '12 at 13:42)
Kirk Roberts
I don't disagree that someone can't contribute without a CS or electronics background, but every implementation of AI is going to require someone capable of implementing the ideas -- that's why it's traditionally associated with computer science. Furthermore, it's generally something people study in-depth as part of a graduate course. You don't usually have to have studied CS/electronics in your undergrad to do AI work for your grad degree, but if you want to have any hope of demonstrating the value of your contributions you'll be forced to learn how to implement your ideas.
(Feb 12 '12 at 14:03)
Brian Vandenberg
|
|
Basically the answer is the history of the two fields. People like Turing started applying CS-like ideas to AI. Now we apply ideas from statistics and other mathematical fields, as well as social fields like psychology and linguistics. These are certainly not computer science. But at least in the beginning artificial intelligence was a motivating force behind computer science and the two fields have bootstrapped off each other ever since. CS is where the AI experts concentrated, and the easiest way to be exposed to AI is to be a CS student. +1 So if I understand what you're saying correctly, you're saying it was basically like an accident in history? And when you say that people like Turing started applying "CS-like ideas to AI", could you explain which "CS like ideas" fueled the early development of AI?
(Feb 04 '12 at 19:33)
gangsta84
One idea is thinking of intelligence as a black box with some kind of input and output (like a turing machine, and also like the turing test). Another idea is thinking about computational limits and heuristic strategies which match what is known about people well enough.
(Feb 04 '12 at 19:45)
Alexandre Passos ♦
1
Accidental isn't quite how I'd describe it. What we currently call "CS theory" largely happened well after Turing. Cook's paper on NP completeness, for example, wasn't published till 1971. Before that the bounds of computer science were more murky, especially since most of it was theoretical anyhow due to the lack of substantial computing power. And since computers were the first (and to date probably still the only) practical means of implementing an artificial intelligence, it was a logical place for AI to be. As for more examples, one can look to how many AI problems are formulated with classical CS data structures and algorithms. One of the first things you learn in an AI class, for instance, is the classical search mechanisms on graphs (e.g., DFS, BFS).
(Feb 04 '12 at 22:28)
Kirk Roberts
Great answer. The OP makes a fair point that there's much shared between AI and other fields, as the study of "thought" in general is being pursued from many angles. AI, machine learning, cognitive science, psychology, neuroscience, etc. They're separate because they're all relatively young sciences. The more they mature, the more they'll merge together.
(Feb 05 '12 at 21:45)
Cerin
I'd expand to say there's no reason that AI couldn't be dominated by the field of philosophy, with computers used, but viewed as one tool along with theory and more traditional introspective approaches. Judea Pearl was a philosopher, after all, and his work forms the basis of a HUGE portion of modern ML work. Thinking about thinking and logic-based approaches are also within the domain of philosophy.
(Feb 12 '12 at 02:04)
Jacob Jensen
|
Because often the same kind of person who is interested in the other big areas of computer science is interested in AI.
And I'm sure a lot of those people are also interested in physics, no? Then how come physics is considered just physics and not CS?
In practice it's a larger fraction of computer scientists than of physicits (just go to any AI conference, or look up the affiliations of any AI journal, or see which departments offer undergrad AI courses)
I don't think you've clarified anything for the situation. The question that Kaitlyn is asking is why "CS departments are the ones to offer undergrad AI courses", and not some other department....
My opinion is that this is in part an accident of history and in part a consequence of the fact that computer science, in practice, is not "the study of the properties of computation" but "the study of computers and stuff you can do with them", so the idea of trying to do "intelligence" with computers felt natural.